MASTER THESIS Video Watermarking - Computer Graphics Group ...

ecompression and noising, and to some removal attacks such as denoising and
collusion by averaging.
Noise method is the most resistant method to scaling. The watermark is
successfully detected even if the video is scaled down up to 1/5 of the former
resolution. Using any of the frequency domain methods, the watermark is
destroyed when scaling down the video to 1/3 of the former resolution.
On the other hand, noise method is the most vulnerable method to cropping.
The frequency domain methods withstand cropping the video up to 1/5 of the
former resolution while the watermark may be severely impaired by cropping the
video to 1/4 of the former resolution in case of noise method.
Concerning blurring, the noise method watermark is robust using any size
of the convolution blur mask. In case of both block and coefficient methods,
blurring with the 7×7 mask may destroy the watermark but the video quality is
severely degraded as well.
Sharpening even increases the watermark detection success probability in
all the methods.
When using the frequency domain methods, the multiple watermark
embedding test has shown that limited number of transform coefficients enables
overwriting of previously embedded watermarks. Thereby, an attacker may
completely destroy the former watermark.
On the other hand, the frequency domain methods are more resistant to
collusion by swapping macroblocks. Anyway, the watermark may be destroyed
with sufficient number of copies participating in the collusion attack.
Although noise method is more vulnerable to cropping, it is equally or more
resistant to the other attacks than the frequency domain methods. Moreover,
there is only few visual data left in the video cropped to 1/4 of the former
resolution.
Further, the noise method watermark is the least perceptible one in
comparison with the other method watermarks.
Unfortunately, there is a trade-off between benefits and much more bit-rate
growth when using noise method. Noise method increases the video bit-rate up to
two times more than the other methods, using reasonable weight factors. The
ratio grows up with increasing weight factor values.
With respect to the reasons above, noise watermarking method is
recommended despite the bit-rate growth. Further, weight factor of 2 is
recommended as a good compromise between robustness and perceptibility.
For practical use, several improvements should be made.
Firstly, the embedding process should be optimized to preserve the former
bit-rate of the video sequences.
51